LLaMA

LLaMA is a dense decoder-only Transformer released for research as a general foundation model. It comes in multiple sizes so labs can study quality versus compute and run smaller variants on modest hardware. Despite modest parameter counts, it showed strong zero and few shot ability and stable context use for its time. Since alignment was minimal, most real deployments fine tuned it for instruction following, safer behavior, and JSON formatting. The release sparked a large ecosystem of derivatives for chat, multilingual work, domain adaptation, and retrieval-augmented apps, making LLaMA a common starting point for efficient, self-hosted assistants.